Custom Models

What custom models are for

A custom model defines reusable assistant behavior. It combines an LLM, a system instruction, and visibility rules.

What can be configured

A custom model can include:

• a name
• a description
• a category
• tags
• an LLM
• a default system instruction
• a default reasoning effort when the LLM supports native reasoning
• visibility
• team access
• a credential mode

When to use a custom model

A custom model is useful when the goal is to:

• keep a stable tone and behavior
• reuse the same instruction across several flows
• prepare a building block that will later be used inside a group model

Visibility and sharing

A model can be:

• private
• shared with teams
• public

Public visibility should be used only for models that are intentionally designed for broader reuse.

BYOK and the vault

When BYOK is used, the model can rely on a secret stored in the Vault. This makes it possible to use custom credentials without exposing them in the model configuration.

If a public model uses BYOK, each user or team must configure an appropriate vault key before using it. The full key stays in the vault and is never displayed in the model page.

Use a custom model

From the custom model list or detail page, Use starts a chat with that model. A custom model target runs in Normal mode.

You can also select Add this agent to add the custom model to an existing group, either as an Agent or as the Brain.

Attach resources

The detail page includes model files. Use them for durable context that should follow the model across chats.

Good practices

• Keep the system instruction short and precise.
• Use the description to explain the intended use case.
• Use clearly differentiated names when several variants coexist.
• Review visibility carefully before making a model public.
• Check BYOK secrets before sharing a model broadly.